Hansa and Cradle: buying AI is the easy part. Changing the workflow is harder.

The October 1 announcement

Hansa Biopharma and Cradle announced a partnership to use AI-driven protein design and engineering across Hansa’s discovery pipeline, with a focus on novel immunomodulatory therapies. The announcement describes their intended use of the platform to design and optimize therapeutic candidates.

A partnership announcement establishes a planned direction. It does not establish the size of any productivity improvement or the success of future candidates. For founders evaluating a similar purchase, the useful question is what must change inside the team for the platform to become part of everyday scientific work.

Our take: adoption happens at a decision boundary

An AI platform can be technically available while the team continues to make decisions through its old workflow. Scientists may try the tool, export a result and then reconstruct its context in a spreadsheet or presentation. The platform has been used, but the handoff has not improved.

We would choose a specific decision boundary for the first deployment: a design review, a candidate selection or a follow-up experiment. Identify who proposes, who reviews and what evidence must be available. Then design the platform’s role around that event. This makes the intended change observable.

The pilot should end with a decision about the workflow, as well as a result from the model.THE BIOMEDANA LENS

A hypothetical pilot: one optimization round

Imagine a team evaluating AI suggestions for a protein optimization round. Before generating variants, it defines the properties that matter, the assays it can run and how conflicting results will be handled. The first review should ask whether the suggestions are testable and whether the team can trace why they were proposed.

After testing, the same pilot should examine the return path. Can the scientist connect each result to the correct design version? Can a reviewer see which objective improved and which remains unresolved? Can the next round reuse the evidence without rebuilding the history? These questions turn a demonstration into an operational evaluation.

The scenario is illustrative. It describes how we would structure a pilot, not the implementation or results of the Hansa–Cradle partnership.

The founder questions that deserve time before rollout

Founders often focus on a model’s capabilities because those are easiest to demonstrate. The implementation questions are less glamorous, but they determine whether a team can use the capability repeatedly.

• Who owns the scientific acceptance criteria and resolves disagreements?
• How will candidate identities, experiments and model versions stay connected?
• What can the team export when a program, partner or platform changes?
• How are source evidence, interpretations and recommendations distinguished?
• What work will scientists stop doing if the new workflow succeeds?
• Which failure would make the team narrow or discontinue the pilot?

The tradeoff: local usefulness versus wider reuse

A lightweight pilot can deliver value quickly because it serves one team’s immediate needs. The risk is that its naming, assumptions and evidence format become hard to reuse elsewhere. Conversely, designing an organization-wide data architecture before the first decision can delay learning about what users actually need.

We would keep the scientific scope narrow while making identity, provenance and export explicit from the start. These are small commitments with consequences beyond the pilot. They let a useful local workflow become part of a wider development process without requiring the entire organization to change at once.

Measure the change in work, then decide whether to expand

Look at how long it takes to prepare a review, how often reviewers request missing context and whether experimental results can be reused in the next design round. Include the time spent entering and correcting data. A workflow that saves analysis time but creates an equivalent reconciliation task elsewhere may need redesign.

We would also interview the scientists doing the handoff. Ask them to reconstruct a decision from a previous round using only the retained record. If they need a particular colleague to remember why a candidate advanced, there is still an opportunity to improve the workflow.

The BioMedAna perspective is that adoption should be judged by supported scientific decisions and reusable evidence. October’s partnership news is a reminder that AI is entering operational workflows. For a founder, the work begins with choosing exactly where it should help.

Who owns the extra work a pilot creates?

New tools can shift work between roles. A scientist may save time generating designs while another person spends longer matching the output to material records or correcting a result import. If the evaluation counts only the first task, it can report an improvement while the workflow as a whole becomes harder to operate.

We would map the work before the pilot, including preparation, transfer, review and correction. Then ask participants to record where the new process changes those tasks. The aim is to discover the actual cost of adoption while there is still time to revise the design. This does not require a burdensome time study; a consistent account of recurring friction is a useful start.

Assign an owner to each new responsibility. If a platform needs better input structure, somebody must decide how that structure is maintained. If exceptions require review, somebody must have the context and authority to resolve them. These operational choices deserve a place in the pilot plan.

What we would pass on to another builder

Set an expansion criterion before the first successful demonstration. Specify what the pilot must show about scientific usefulness, evidence handling and the work required from the team. This helps prevent enthusiasm for an attractive output from replacing the original evaluation question.

Test portability while the scope is small. Export one completed round and ask whether a colleague can understand the candidate identities, predictions, experiments and review outcome outside the primary interface. The question is whether the team retains a usable scientific record as its partners and tools change.

Give scientists permission to identify where the platform is unhelpful. A narrowly useful workflow can still be a successful pilot if its limits are understood. The founder decision is whether to improve that workflow, expand it to a similar use or choose another problem. The record of the pilot should make that decision easier to defend.

Give the pilot a decision it must earn

A bounded pilot makes the adoption question easier to evaluate. Agree on the work that should change and the evidence needed to justify expansion.

The questions behind the headline

Follow the evidence

1. Cradle — Hansa Biopharma partnership for AI-driven protein design and engineering (October 1, 2026) ↗

Original BioMedAna editorial analysis. The linked sources support the reported developments; examples, frameworks and product implications reflect our perspective.

Explore the underlying concepts

• Bioprocess AI platform evaluation
• BioMedAna Assist & Agents